一、实验数据准备

我们使用的是MIT67数据集,这是一个标准的室内场景检测数据集,一个有67个室内场景,每类包括80张训练图片和20张测试图片 读者可通过以下网址下载

但是数据集较大,下载花费时间较长,所以建议私信我发给你们

数据集

将下载的数据集解压,主要使用Image文件夹,这个文件夹一共包含6700张图片,还有它们标签的txt文件

大体流程分为以下几步

二、数据预处理和准备

1:数据集的读取

2:重载data.Dataset类

3:transforms数据预处理

三、模型构建

1:ResNet-50网络

网络结构图如下

PyTorch搭建卷积神经网络(ResNet-50网络)进行图像分类实战(附源码和数据集)-LMLPHP

2:bottleneck的实现

结构图如下

PyTorch搭建卷积神经网络(ResNet-50网络)进行图像分类实战(附源码和数据集)-LMLPHP

 

3:ResNet-50卷积层定义

4:forward函数的实现

5:预训练参数装载

四、模型训练与结果评估

1:训练类的实现

2:优化器的定义

3:学习率衰减

4:训练

训练过程如下

PyTorch搭建卷积神经网络(ResNet-50网络)进行图像分类实战(附源码和数据集)-LMLPHP

 最后 部分代码如下

1.py

import torch
from torch.autograd import Variable as V
import torchvision.models as models
from torchvision import transforms as trn
from torch.nn import functional as F
import os
import numpy as np
from mymodels import *
from PIL import Image
import torch.utils.data as data
from torch.utils.data import DataLoader
from utils import *
tmp_dir = '/home/yyh/tmpmit67'
import torch.optim as optim
import matplotlib.pyplot as plt
import time
import json
def get_im_list(im_dir, file_path):
    im_list = []
    im_labels = []
    im_origin = []
    with open(file_path, 'r') as fi:
        for line in fi:
            im_list.append(im_dir + line.split()[0])
            im_labels.append(int(line.split()[-1]))
            im_origin.append(line.split()[0])
            array = line.split('/')
    return im_list, im_labels, im_origin
ate(fi):
            sname = line.strip()
            sdict[sid] = sname
    return sdict
_sdict = sun397_sdict()


arch = 'resnet50'
# load the pre-trained weights
model_file = '%s_places365.pth.tar' % arch
if not os.access(model_file, os.W_OK):
    weight_url = 'http://places2.csail.mit.edu/models_places365/' + model_file
    os.system('wget ' + weight_url)
model = resnet50(num_classes=365)
checkpoint = torch.load(model_file, map_location=lambda storage, loc: storage)
state_dict = {str.replace(k,'module.',''): v for k,v in checkpoint['state_dict'].items()}
model.load_state_dict(state_dict)
model.fc = torch.nn.Linear(2048,67)
model.eval()




"""
model = resnet50(num_classes=67)
pretrained = torch.load("/home/yyh/fineTune/mit67_place/model_epoch_30.pth").module
state_dict = pretrained.state_dict()
model.load_state_dict(state_dict)
model.eval()

"""





# load the image transformer
transform_train = trn.Compose([
        trn.Scale(256),
        trn.RandomSizedCrop(224),
        trn.RandomHorizontalFlip(),
        trn.ToTensor(),
        trn.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
transform_test = trn.Compose([
        trn.Scale(256),
        trn.CenterCrop(224),
        trn.ToTensor(),
        trn.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
# load the class label

def default_loader(path):
    return Image.open(path).convert('RGB')
class MyDataset(data.Dataset):
    def __init__(self, images, labels,loader=default_loader,transform=None):
        self.images = images
        self.labels = labels
        self.loader = loader
        self.transform = transform
    def __getitem__(self, index):
        img, target = self.images[index], self.labels[index]
        #print(img)
        img = self.loader(img)
        if self.transform is not None:
            img = self.transform(img)
        #print(img)
        return img, target
    def __len__(self):
        return len(self.images)

imdir = r'C:\Users\Admin\Desktop\MIT67\Images/'
train_file = 'C:\Users\Admin\Desktop\MIT67\TrainImages.label'
test_file = 'C:\Users\Admin\Desktop\MIT67\TestImages.label'
#train_file = test_file
train_list, train_labels,img_path= get_im_list(imdir, train_file)
test_list, test_labels ,img_path_2= get_im_list(imdir, test_file)
batch_size = 16
net = model
net.cuda()



#print(test_js)


for i in range(0, len(train_list)):
    path = img_path[i]
    save = []
    print(path)
    json_name = (path.replace("/", "_")).replace(".jpg", ".json")
    f_train = open("C:\Users\Admin\Desktop\rgbd_data\annotated_area/" + json_name)
    train_js = json.load(f_train)
    if len(train_js) == 0:
        train_js.append( {"classname":"unknown","bbox":[0,0,223,223],"score":1})
    for j in range(0, len(train_js)):
        data, target = train_list[i], train_labels[i]
        data = Image.open(data).convert('RGB')
        json_data = train_js[j]["bbox"]
        data = data.resize((224, 224), Image.ANTIALIAS)
        print(json_data)
        data = data.crop([json_data[0], json_data[1], json_data[2], json_data[3]])
        data = data.resize((224, 224), Image.ANTIALIAS)
        data = transform_test(data)
        newdata = torch.zeros(1, 3, 224, 224)
        newdata[0] = data
        data = Variable(newdata).cuda()
        output, record = net(data)
        data = record.cpu().detach().numpy()
        save.append(data)

    data = save[0]
    for j in range(1, len(train_js)):
        data += save[j]
    data = data / len(train_js)
    # print(data)
    # target = Variable(target).cuda()

    # print(output)
    # print(output["avgpool"].cpu().shape)
    root = "/home/yyh/PycharmProjects/feature_extractor/loc_224_npy/" + path.split("/")[0]
    if not os.path.exists(root):
        os.makedirs(root)
    dir = "/home/yyh/PycharmProjects/feature_extractor/loc_224_npy/" + path.replace(".jpg",".npy")
    np.save(dir, data)
    print(i)


for i in range(0, len(test_list)):
    path = img_path_2[i]
    save = []
    print(path)
    json_name = (path.replace("/", "_")).replace(".jpg", ".json")
    f_test = open("/home/yyh/rgbd_data/annotated_area/" + json_name)
    test_js = json.load(f_test)
    if len(test_js) == 0:
        test_js.append( {"classname":"unknown","bbox":[0,0,223,223],"score":1})
    for j in range(0, len(test_js)):
        data, target = test_list[i], test_labels[i]
        data = Image.open(data).convert('RGB')
        json_data = test_js[j]["bbox"]
        data = data.resize((224, 224), Image.ANTIALIAS)
        print(json_data)
        data = data.crop([json_data[0], json_data[1], json_data[2], json_data[3]])
        data = data.resize((224, 224), Image.ANTIALIAS)
        data = transform_test(data)
        newdata = torch.zeros(1, 3, 224, 224)
       tach().numpy()
        save.append(data)


    data = save[0]
    for j in range(1, len(test_js)):
        data += save[j]
    data = data / len(test_js)
    print(data)
    root = "/home/yyh/PycharmProjects/feature_extractor/loc_224_npy/" + path.split("/")[0]
    if not os.path.exists(root):
        os.makedirs(root)
    dir = "/home/yyh/PycharmProjects/feature_extractor/loc_224_npy/" + path.replace(".jpg",".npy")
    np.save(dir, data)
    print(i)

    #time.sleep(10)








#print(net)
#train_net = torch.nn.DataParallel(net, device_ids=[0])
#optimizer = optim.SGD(params=train_net.parameters(), lr=0.001, momentum=0.9, weight_decay=1e-4)
#scheduler = StepLR(optimizer, 30, gamma=0.1)
#trainer = Trainer(train_net, optimizer, F.cross_entropy, save_dir="./mit67_imagenet_448")
#trainer.loop(130, train_loader, test_loader, scheduler)

2.py

from pathlib import Path
import torch
from torch.autograd import Variable
from torch.optim import Optimizer
from torch import nn
from tqdm import tqdm
import torch.nn.functional as F

    expansion = 4

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(Bottleneck, self).__init__()
        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(planes)
        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,
                               padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(planes)
        self.conv3 = nn.Conv2d(planes, planes * self.expansion, kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(planes * self.expansion)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn self.conv3(out)
        out = self.bn3(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.relu(out)
        return out


    def __init__(self, block, layers, num_classes=1000):
        self.inplanes = 64
        super(ResNet, self).__init__()
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
                               bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
        self.avgpool = nn.AvgPool2d(kernel_size=7, stride=1, padding=0)
        self.fc = nn.Linear(512 * block.expansion, num_classes)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
            elif isant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.inplanes, planes * block.expansion,
                          kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(planes * block.expansion),
            )

        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes * block.expansion
        for i in range(1, blocks):
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)

    def forward(self, x):
        record = dict()
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        record["maxpool"] = x
        x = self.maxpool(x)
        x = self.layer1(x)
        record["layer1"] = x
        x = self.layer2(x)
        record["layer2"] = x
        x = self.layer3(x)
        record["layer3"] = x
        x = self.layer4(x)
        record["layer4"] = x
        x = selpool"] = x
        x = self.fc(x)
        return x,record["avgpool"]


def rined=False, **kwargs):
    """Constructs a ResNet-50 model.
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)
    return model

3.py

from pathlib import Path
import torch
from torch.autograd import Variable
from torch.optim import Optimizer
from torch import nn
from tqdm import tqdm


a.is_available()
    torch.backends.cudnn.benchmark = True

    def __init__(self, model, optimizer, loss_f, save_dir=None, save_freq=10):
        self.model = model
        if self.cuda:
            model.cuda()
        self.optimizer = optimizer
        self= save_dir
        self.save_freq = save_freq

    def _iteration(self, data_loader, is_train=True):
        loop_loss = []
        accuracy = []
        for data, target in tqdm(data_loader, ncols=80):
            if self.cuda:
                data, target = data.cuda(), target.cuda()
            output = self.model(data)
            loss = self.loss_f(output, target)
            loop_loss.append(loss.data.item() / len(data_loader))
            accuracy.append((output.data.max(1)[1] == target.data).sum().item())
            if is_train:
                self.optimizer.zero_grad()
                loss.backward()
                self.optimizer.step()
        mode = "train" if is_train else "test"
        #print(">>>[{}] loss: {:.2f}/accuracy: {:.2%}").format(mode,sum(loop_loss),float(sum(accuracy)) / float(len(data_loader.dataset)))
        print(mode)
        print(sum(loop_loss))
        print(float(sum(accuracy)) / float(len(data_loader.dataset)))
        
        return loop_loss, accuracy

    def train(self, data_loader):
        self.model.train()
        with torch.enable_grad():
            loss, correct = self._iteration(data_loader)

    def test(self, data_loader):
        self.model.eval()
        with torch.no_grad():
            loss, correct = self._iteration(data_loader, is_train=False)

    def loop(self, epochs, train_data, test_data, scheduler=None):
        for ep in range(1, epochs + 1):
            if scheduler is not None:
                scheduler.step()
            print("epochs: {}".format(ep))
            self.train(train_data)
            self.test(test_data)
            if ep % self.save_freq == 0:
                self.save(ep)

    def save(self, epoch, **kwargs):
        if self.save_dir is not None:
            model_out_path = Path(self.save_dir)
            state = self.model
            if not model_out_path.exists():
                model_out_path.mkdir()
            print(self.save_dir+ "model_epoch_{}.pth".format(epoch))
            torch.save(state, self.save_dir+ "/model_epoch_{}.pth".format(epoch))


class _LRScheduler(object):
    def __init__(self, optimizer, last_epoch=-1):
        if not isinstance(optimizer, Optimizer):
            raise TypeError('{} is not an Optimizer'.format(
                    type(optimizer).__name__))
        self.optimizer = optimizer
        if last_epoch == -1:
            for group in optimizer.param_groups:
                group.setdefault('initial_lr', group['lr'])
        else:
            for i, group in enumerate(optimizer.param_groups):
                if 'initial_lr' not in group:
                    raise KeyError("param 'initial_lr' is not specified "
                                   "in param_groups[{}] when resuming an optimizer".format(i))
        self.base_lrs = list(map(lambda group: group['initial_lr'], optimizer.param_groups))
        self.step(last_epoch + 1)
        self.last_epoch = last_epoch

    def get_lr(self):
        raise NotImplementedError

    def step(self, epoch=None):
        if epoch is None:
        tep_size, gamma=0.1, last_epoch=-1):
        self.step_size = step_size
        self.gamma = gamma
        super(StepLR, self).__init__(optimizer, last_epoch)

    def get_lr(self):
        return [base_lr * self.gamma ** (self.last_epoch // self.step_size)
                for base_lr in self.base_lrs]

创作不易 觉得有帮助请点赞关注收藏~~~

11-11 10:28